Research map / Architectures and efficiency / Neural architecture search
SMASH: One-Shot Model Architecture Search through HyperNetworks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the CIFAR-100-found architecture rather than the STL-10-found architecture | SMASH architecture search and transfer to STL-10 · STL-10 (full training set) | error rate: 20.275% -> 17.54% | Helped |
| Use one main convolution with 3x3 filters (ablation) | SMASHv2 · STL-10 (10-fold training splits) | error rate: 41.52 ± 2.10 -> 37.76 ± 0.58 | Helped |
| Reduce HyperNet capacity (ablation) | SMASH architecture scoring · CIFAR-100 architecture ranking | correlation between SMASH score and true validation performance | Hurt |
| Increase the proportion of freely learned parameters relative to HyperNet-generated parameters (ablation) | SMASH architecture scoring · architecture ranking | correlation between SMASH score and true performance | Hurt |
| Give the HyperNet the correct architecture encoding instead of a corrupted encoding (ablation) | SMASH · architecture validation scoring | validation performance | Helped |
| Use gradient-directed perturbations instead of random architecture perturbations (ablation) | SMASH architectural search · architecture search | SMASH score | No clear effect |
| Initialize the resulting network with SMASH-generated weights instead of standard initialization (ablation) | SMASH-derived resulting network · resulting-network training | training performance | Hurt |
Closest papers by meaning
- Cream of the Crop: Distilling Prioritized Paths For One-Shot Neural Architecture Search (NeurIPS 2020 · similarity 0.89)
- Graph HyperNetworks for Neural Architecture Search (ICLR 2019 · similarity 0.89)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.89)
- HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning (AAAI 2025 · similarity 0.87)
- Principled Architecture-aware Scaling of Hyperparameters (ICLR 2024 · similarity 0.87)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.